|
PureBytes Links
Trading Reference Links
|
How are you setting up a system in MS to study options trading?
Daryl Roberts
-----Original Message-----
From: owner-metastock@xxxxxxxxxxxxx
[mailto:owner-metastock@xxxxxxxxxxxxx]On Behalf Of rudolf stricker
Sent: Friday, January 15, 1999 5:52 AM
To: metastock@xxxxxxxxxxxxx
Subject: how to improve generalization in system optimization
Working on a trading system for options, I got nice results in terms
of "consistency of profit" by combining different criterions during
the optimization process, such as
profit,
number of winning / number of loosing trades,
sum of wins / sum of losses, and
shape of the loss distribution.
Even if the profit of the system in a given time frame was reduced by
a factor of up to 10 by using these combined criterions, consistency
of the results ensures, that drawbacks are small and money management
becomes an easy task, because of the "well-behaved" loss
distributions.
But still I'm not satisfied with the generalization capabilities of
the system, i.e. its out-of-sample results. To improve generalization,
I use some techniques, like
periodical (e.g. weekly) adaptive refinement of the system
and increasing weight function for "newer" trades.
My questions on this topics are:
What is a "good generalization result" for a trading system? Is it for
instance a good result to get an out-of-sample result which is about
50% of the result from system optimization?
Are there other important techniques (beside of periodical
re-optimization and weighting along time) to improve generalization
capabilities of a system? - Where should I look at? - What should I
try?
Is there also a risk for overdoing the adaptive refinement and / or
the weighting along time? - What could be seen as a relevant criterion
to find the right measure, e.g. for the "update frequency" and the
ratio of maximum / minimum weight?
Any helpful hint is welcome!
mfg rudolf stricker
| Disclaimer: The views of this user are strictly his own.
|